DON'T. All money that can be pumped out of the market by standard ML algorithms has been pumped out by standard ML algorithms.
Ask HN: Has anyone used machine learning to trade?
41–50 of 64 posts
Re: Ask HN: Has anyone used machine learning to trade?
#42Don't bother.
D.E. Shaw, Renaissance Tech, Goldman Sachs, etc. have already hired hundreds (if not thousands) of people smarter than you to do exactly this. They have more Math, Physics, Econ, CS PhDs than you can count/
Re: Ask HN: Has anyone used machine learning to trade?
#43Earlier quoted context omitted.
A strategy: run a LOT of simulations, keep running them every day. When they are too good to ignore, go raise some capital.
people run MC all the time, not really a new strategy -- expect no breakthrough on that...
I'm suggesting that you run all of your trades with play money, and see how it goes. I'm not talking about Monte Carlo simulations.
Re: Ask HN: Has anyone used machine learning to trade?
#44Earlier quoted context omitted.
Reality begs to differ.
If by reality you mean Buffett or Soros, consider the thousands (hundreds of thousands?) of extremely smart and hardworking people who fail to replicate their success.
By reality one could consider any of a number of small prop trading shops which do rather well.
Re: Ask HN: Has anyone used machine learning to trade?
#45Noticing comments on the high degree of proficiency of the competition(or so is the perception.) Anyone know of any particular markets where a small timer might have a better go at it due to insufficient volume to attract the bigger players, or for any other factors?
If anyone knew, they would be crazy to tell you.
Besides, the bigger players could easily afford to risk a few hundred grand finding out whether they can plug the existing system into the small market profitably.
Re: Ask HN: Has anyone used machine learning to trade?
#46Can't you just download a historical set and use that to test whatever algorithms you think you would use? Then when you are confident enough in their performance, you can take the risk? I wouldn't say it would be a complete waste of time because you would be sure to learn something. If you think you can realistically make money--well then, that's a different question that I wouldn't be fit to answer.
> Can't you just download a historical set and use that to test whatever algorithms you think you would use? Be careful - much of the historical data has been pruned to eliminate some data from companies that didn't survive. I've heard that one of the major quant houses (Shaw?) can often guess what data set you used to test your strategy by looking at the strategy. Plus, there's a selection or fitting problem. Your s…
You make some good points, but isn't this always going to be a problem when training your algorithms? You can't really train in real time so by the nature of the industry you're always going to be put at a disadvantage by using older data.
Re: Ask HN: Has anyone used machine learning to trade?
#47The majority of people who have responded to you with various pieces of opinion/information forgot to answer this question:
Does anyone have any experience in doing this?
So take everything with a grain of salt-- there seems to be a lot of advice going around by people who don't have skin in the game-- especially those who say it can't be done. Unless they've put money down to test this, then their opinion should be discounted.
For those saying commissions are to high for a shorter term timeframe: trading has been commoditized. Lightspeed offers .40 per lot, and IBKR offers around that given a certain amount of trading volume. Forex brokers make money off the spread, and so the commission that you pay is only from the bid ask. Since you are trading small amounts, you will get hit on the comission structure, but if you pay per share rather than per trade it won't hurt as bad. You also might want to look into equity futures as well-- very liquid, good tax advantages, and good commission structures.
Biggest thing about small sizes is that you're subject to the pattern daytrader rule-- you can't make >3 intraday trades if your portfolio is I'm a discretionary trader, and not a system trader. I am currently developing some systems to go live but they are not ready as of yet.
I also know of many who are quants, and it's possible and profitable.
Re: Ask HN: Has anyone used machine learning to trade?
#48Earlier quoted context omitted.
Reality begs to differ.
If by reality you mean Buffett or Soros, consider the thousands (hundreds of thousands?) of extremely smart and hardworking people who fail to replicate their success.
Since we're talking about quants and not value/macro, then Buffett and Soros have little to do with this.
Re: Ask HN: Has anyone used machine learning to trade?
#49Don't bother. D.E. Shaw, Renaissance Tech, Goldman Sachs, etc. have already hired hundreds (if not thousands) of people smarter than you to do exactly this. They have more Math, Physics, Econ, CS PhDs than you can count/
They have more Math, Physics, Econ, CS PhDs than you can count/
So did BSC, AIG, LEH, and MER
Re: Ask HN: Has anyone used machine learning to trade?
#50intraday NN is irrelevant because NN are prone to overlearning, so the stuff becomes even more horrible - actually in reality you don't care which way it moves, as long as it moves: this is why the "big fish". It is like owning the place in Vegas: you don't care about the bets (much less about their substance or "direction"), as long as there are bets. if you really want intraday, think about news.google.com and http…
intraday NN is irrelevant because NN are prone to overlearning
I call bullshit. A lot of the most successful applications in machine learning are modeled using neural networks.
To regularize a neural network, you use standard machinery that controls the capacity of the machine. The scale of the weights cannot be too large (this is a famous result from IIRC Haussler in 1992) so penalize the weights using an l2 (Gaussian) prior. This reduces the model complexity and the potential for overfitting.